Application of Gas Chromatography− Mass Spectrometry Metabolite

although recent developments in analytical techniques and data analysis have made possible the detailed analysis of minor components that may infl...
0 downloads 0 Views 257KB Size
J. Agric. Food Chem. 2007, 55, 1129−1138

1129

Application of Gas Chromatography−Mass Spectrometry Metabolite Profiling Techniques to the Analysis of Heathland Plant Diets of Sheep IFAT PARVEEN,*,† JON M. MOORBY,† MARIECIA D. FRASER,† GORDON G. ALLISON,† AND JOACHIM KOPKA‡ Institute of Grassland and Environmental Research, Plas Gogerddan, Aberystwyth SY23 3EB, United Kingdom, and Max Planck Institute of Molecular Plant Physiology, Am Mu¨hlenberg 1, D-14467 Golm, Germany

Little is known about how plant biochemistry influences the grazing behavior of animals consuming heterogeneous plant communities. The biochemical profiles of grassland species are mostly restricted to major nutritional characteristics, although recent developments in analytical techniques and data analysis have made possible the detailed analysis of minor components that may influence animal feeding preferences, performance, and health. In the present study, gas chromatography coupled with time-of-flight mass spectrometry (GC-TOF/MS) was used to profile the abundances of metabolites in nine specific heathland plant groups and in three mixed forage diets containing 10, 20, or 30% heather (Calluna vulgaris) and also in plasma and feces from sheep offered one of the three diets. Statistical and chemometric approaches, that is, principal component analysis (PCA) and hierarchical cluster analysis (HCA), were used to discriminate between these diets and between individual animals maintained on these diets. It is shown that GC-TOF/MS analysis of sheep plasma allowed distinction between the very similar diets by PCA and HCA, and, moreover, the plant metabolites responsible for the differences observed have been identified. Furthermore, metabolite markers of herbage mixtures and individual plant groups have been identified, and markers have been detected in sheep plasma and feces. KEYWORDS: Diet composition; GC-TOF/MS; ruminant nutrition; metabolite markers

INTRODUCTION

Free-ranging ruminants are significant components of the agriecosystem, and their health and production performance depend upon the nutritive value of the complex plant communities available for consumption. Choices made by large herbivores regarding the type and quantity of plant material grazed can have a profound effect on species richness and diversity and, consequently, on the structure and function of the agriecosystem. Furthermore, the dynamic distribution of nutrients and minerals through trampling and excretion can also affect the ecosystem (1). Therefore, a key element to understanding factors affecting long-term sustainability of ecosystems is an understanding of foraging preferences on heterogeneous swards. This would provide objective guidance for efficient range management through habitat restoration and maintenance and the development of management guidelines for grazing sensitive ecosystems. Many plants consumed by herbivores contain the nutrients needed to meet basic requirements, but they can also contain a * Corresponding author [telephone +44 (0) 1970 823207; fax +44 (0) 1970 828357; e-mail [email protected]]. † Institute of Grassland and Environmental Research. ‡ Max Planck Institute of Molecular Plant Physiology.

diverse and complex array of secondary compounds that provide some degree of defense against predation, disease, competition, and adverse climatic conditions. These compounds are frequently antinutritional or toxic, yet little is known about how they influence the choices and aversions of animals in their grazing behavior and the associated impact on health and performance. In addition, the composition of primary metabolites varies greatly among sward components and has significant effects on animal performance and dietary selection. Few detailed studies have been conducted with free-ranging animals grazing complex plant communities. This is primarily due to difficulties in accurately determining animal intake and diet composition in such environments. Earlier investigations involved direct observation of the grazing animal (2, 3). However, although the method proved to be simple, problems in species identification and quantification of plants consumed were major disadvantages. More recent approaches include microhistological procedures (1, 3-5), stable C-isotope discrimination (1, 6-8), use of plant wax marker compounds (1, 9-11), and near-infrared spectroscopy (NIRS) (12-15). The advent of the postgenomic era has brought powerful highthroughput analytical methods coupled to advanced chemometric techniques. One area that has benefited directly from these

10.1021/jf062995w CCC: $37.00 © 2007 American Chemical Society Published on Web 01/24/2007

1130

J. Agric. Food Chem., Vol. 55, No. 4, 2007

advances is metabolite profiling, and various strategies are being used to explore large populations of metabolites in biological fluids (16). Although many of the reported techniques allow identification of both ruminant diet and botanical composition, the use of gas chromatography with time-of-flight mass spectrometry (GC-TOF/MS) as a profiling tool permits identification of metabolites within samples. Whereas alternative GCMS technologies, for example, quadrupole MS, would also allow the profiling of metabolites in complex samples, TOF/MS, a nonscanning technology, is highly reproducible and allows much higher rates of data acquisition, which increases analytical resolution and permits more rigorous mass spectral deconvolution and metabolite identification. In the present study, we hypothesized that it is possible to correlate the presence or absence of specific compounds in blood plasma or feces with the consumption of specific plants in the diet of sheep. The objective of this study was to use GC-TOF/MS to profile metabolites present in a range of plants at different concentrations in diets fed to sheep and to determine the presence or absence of these metabolites and their derivatives in blood and feces. Furthermore, we report on how information on plant chemical composition can be used to discriminate between animals on each of the different dietary regimens on offer. MATERIALS AND METHODS The experimental design and sampling procedures have been reported in detail elsewhere (17). Briefly, during a zero-grazing experiment, material from two indigenous vegetation communities, a Molinia caerulea dominated grassland (hill grass mix) and a Calluna Vulgaris dominated dwarf-shrub community (heather mix), was offered to sheep in three different ratios: 10% heather mix/90% hill grass mix; 20% heather mix/80% hill grass mix; and 30% heather mix/70% hill grass mix. All ratios were prepared on a fresh matter basis. Each diet was offered to four animals each of a group of 12 mature Welsh Mountain ewes, for a period of 29 days, with 22 days for adaptation and 7 days for measurements. Samples of the sward mixtures were collected daily, bulked over the course of the measurement week, and stored at -20 °C. These bulked samples were subsequently thoroughly mixed and subsampled and separated into categories based on botanical classification. These plant samples were freeze-dried and ground to pass through a 1 mm sieve in preparation for analysis. The amount of feed refused by each animal were recorded daily, and a subsample was taken to determine dry matter content by drying overnight at 100 °C. The remaining refusals were then bulked on an individual animal basis, before being separated to determine botanical composition. This information, together with the botanical composition of the diet as offered, was used to calculate the composition of the diet consumed by each animal. Total fecal output was recorded daily, and a subsample from each animal was retained and stored at -20 °C. At the end of the measurement week the daily fecal samples were bulked on an individual animal basis, thoroughly mixed, and subsampled. This sample was then freeze-dried and milled to pass through a 1 mm sieve in preparation for chemical analysis. A blood sample was collected from each animal at the end of the measurement period. Blood was collected from the jugular vein into Vacuette evacuated collection tubes containing lithium heparin (Greiner Labortechnik, Kremsmu¨nster, Austria) and immediately placed on ice. Within an hour of collection, blood cells were separated from plasma by centrifugation (approximately 1700g for 25 min at 4 °C), and the plasma was transferred in approximately 1 mL aliquots into microcentrifuge tubes for storage at -80 °C. Sample Preparation for Analysis by GC-TOF/MS. Samples were prepared for derivatization to the methoxime/trimethylsilyl (TMS) derivative. Portions of plasma (0.2 mL) were mixed with 0.4 mL of ice-cold acetone and incubated at 4 °C for 1-2 h to precipitate protein. The plasma samples were then centrifuged at 13000g for 5 min in a benchtop centrifuge, and a portion of the supernatant (350 µL) was dried at 40 °C under a stream of nitrogen. Samples of animal feces

Parveen et al. and plant material were prepared for derivatization using an alternative method. Portions of each sample (60 mg) were extracted into 300 µL of ice-cold methanol using a Retsch mill (3 × 2 min). The samples were then mixed with 200 µL of chloroform and incubated with shaking for 5 min at 37 °C, after which a further 400 µL of water was added. The samples were vortexed and then centrifuged for 5 min at 13000g in a benchtop centrifuge to separate the phases. A portion (80 µL) of the upper phase was removed and transferred to a clean microcentrifuge tube and dried in a centrifugal vacuum desiccator until dry at 30 °C. Samples were derivatized for analysis by the addition of 40 µL of freshly prepared methoxyaminehydrochloride in pyridine (20 mg/mL) and incubated in a shaking incubator at 30 °C for 1.5 h. A 10 µL aliquot of a mixture of alkanes (C10-C37) was added to each vial to allow measurement of mass spectral metabolite tag (MST) retention index, and the samples were derivatized to the TMS form by the addition of 70 µL of N-methyl-N-trimethylsilyltrifluoroacetamide (MSTFA) and incubation at 37 °C for 45 min. For sample analysis, a GC 6890 (Agilent Technologies, Palo Alto, CA) was coupled to a time-of-flight (TOF) mass spectrometer (Pegasus II MS system; LECO, St. Joseph, MI). The GC was operated under electronic pressure control and equipped with a split/splitless capillary inlet. Injection was 1 mL in the splitless mode with a 2 min pulse at 110 psi and the injection temperature set to 230 °C. The GC capillary column used was a 30 m × 0.25 mm i.d. Rtx-5Sil MS with an integrated guard column and a 0.25 mm film (Restek GmbH, Bad Homburg, Germany). Helium was used as carrier gas with constant flow at 1 mL/min. The temperature program was 2 min at 80 °C followed by a 15 min ramp to 350 °C and final heating for 2 min at 350 °C. The transfer line to the mass spectrometer was set to 250 °C. The mass spectrometer source was set to 200 °C. Mass spectra were monitored with an acquisition rate of 6 spectra/s in the mass range m/z 70-600. Tuning and all other settings of the mass spectrometer were according to the manufacturer’s recommendations. Mass spectral metabolite tags were obtained by automated deconvolution of GC-MS chromatograms using Chromatof software (LECO) and identified tentatively by automated comparison of mass spectra match and retention time index with deconvoluted MSTs in a public access retention indexed mass spectral library containing 6205 mass spectra (http://csbdb.mpimp-golm.mpg.de/). Peak identification was made by best retention index match hit, and identities greater than (() 20 retention index units were eliminated. Compounds that appeared in three or fewer samples were discarded as erroneous. Peak abundances of compounds occurring as multiple peaks in the chromatogram were combined and exported to Matlab version 6.5.1 (The Mathworks, Cambridge, U.K.), and Genstat version 8.1 (VSN International Ltd., Hemel Hempstead, U.K.) for statistical analysis. Chemometric Data Analysis. Blood plasma GC-MS data were analyzed by principal component analysis (PCA) (18) using the statistics toolbox of Matlab. Compound total ion abundances were standardized to proportion of total ion current to account for differences in sample concentration and injection. Data were standardized for some analyses using the “prestd” function of the neural network toolbox of Matlab. This function mean centered the data from each sample to a standard deviation of one. Principal components (PCs) were calculated for the complete data set on both raw and normalized data. Pairs of PCs were then plotted in a series of ordination diagrams for visual inspection. Where separation between animals in different dietary groups was found, the PC loadings were interrogated to identify compounds that contributed to this separation. High positive or negative loadings indicated the variables that were most influential on the spatial distribution of the samples on the respective ordination diagrams. Furthermore, hierarchical cluster analysis (HCA) was used to estimate linkages between diet classes within the data set. Euclidean distance on the PCs with Ward’s linkage methods was used to derive a similarity matrix, which was processed by agglomerative or divisive clustering algorithms to construct a dendrogram. PCA was also carried out on the fecal GC-MS data as described above. Where separation between the three groups was evident by visual inspection of score plots, HCA was carried out as above. Due to the limited number of plant samples (i.e., one sample per dietary treatment), PCA and HCA were not carried out on the plant metabolite data.

J. Agric. Food Chem., Vol. 55, No. 4, 2007

Metabolite Profiling of Sheep and Their Diets Table 1. Percentage Composition of Diet Components in Each of the Three Dietary Treatments, As Consumed treatment (heather in diet) plant species

10%

20%

30%

Calluna vulgaris (heather) Erica tetralix (bell heather) Molinia caerulea (purple moor grass) Vaccinium myrtillus (bilberry) Juncus effuses (soft rush) Festuca spp. (grass, fescue) Carex spp. (sedge) dead grass (all species) moss (unclassified)

12.3 0.07 40.8 0.16 11.5 1.56 3.71 22.7 7.28

23.8 0.14 35.1 0.31 10.0 1.39 3.20 19.5 6.52

34.6 0.20 29.8 0.44 8.7 1.24 2.72 16.6 5.81

Data and Statistical Analysis. Plant metabolite data were examined in Microsoft Excel to determine those compounds that were predominantly associated with a particular plant group. Compounds that had g70% of their total relative abundance (across all plant groups) associated with one plant group were deemed to be predominantly attributable to that plant group. Pearson’s product-moment correlation coefficients (R) were calculated between metabolites present in the plant, plasma, and feces samples with the amount of heather present in the diet using the “corrcoef” function of Matlab. PCA and HCA were then used to verify that the relative abundances of these metabolites allowed data separation according to dietary regimen. Furthermore, correlations were calculated for the first 10 PC scores of the plasma and feces samples against the total amount of heather eaten and the total amount eaten of each plant subgroup in the mixed diet. Compounds found to be significantly (P < 0.05) correlated with diet were subjected to further investigation by analysis of variance (ANOVA) using Genstat to establish if there were significant differences in the mean abundances of plant components present in the plasma and fecal samples. Treatment effects were partitioned into linear and other effects using polynomial contrasts. RESULTS AND DISCUSSION

The mean compositions of the three experimental diets are presented in Table 1. The mean quantities of C. Vulgaris in the diets were close to the amounts originally specified but differed due to selection of dietary components by individual animals. Relationship of Metabolites in Blood Plasma, Feces, and Plants to the Proportion of Heather in the Diet. Analysis of Plasma Samples. PCA of standardized/normalized plasma data sets showed >99.5% of the total variance was retained by the first 10 components. The first five PCs accounted for >87% of the variance. It was possible to separate samples from animals maintained on the 10% heather diet from samples from animals on the 20 and 30% diets through the combined effect of both PC1 (35.85% of total variance) and PC4 (10.12% of total variance) axes (Figure 1A). Generally, animals maintained on the 20 and 30% diets were separated from animals on the 10% diet along the PC1 axis, whereas there were negative correlations (both P < 0.05) between the percentages of Calluna and Erica in the diet and PC1. Samples taken from animals on the 10% diet had mean positive PC1 scores and negative PC4 scores, whereas samples from animals offered the 20 and 30% diets had mean negative PC1 scores and positive PC4 scores. Examination of the loadings plot (Figure 1B) for these two PCs showed that the 10% diet differed from the other diets in a manner dependent on the concentration of pyroglutamic acid, L-proline, L-isoleucine, 3-aminopiperidin-2-one, and hydrogen cyanamide. Samples from the 20 and 30% diets had lower concentrations of these compounds but greater amounts of malonic acid and L-valine. These two compounds both had high negative PC1 loadings and relatively high positive loadings on

1131

PC4. It was not possible to discriminate between animals on each of the three diets by HCA. Analysis of the plasma data by Pearson’s product-moment correlation indicated other metabolites were significantly correlated (P < 0.05) with diet. Metabolites that were positively correlated with increasing proportion of heather in the diet included L-alanine (R ) 0.61), L-isoleucine (R ) 0.64), L-valine (R ) 0.677), 4-aminobutyric acid (R ) 0.63), 2-propenoic acid (R ) 0.66), 3-methylundecane (R ) 0.69), methoxymethylcyclopropane (R ) 0.65), N,N-diethyl-N′-methyl-1,2-ethanediamine (R ) 0.59), malonic acid (R ) 0.80), succinic acid (R ) 0.73), and galactaric acid (R ) 0.73). In contrast, nicotinamide (R ) -0.58) was negatively correlated with increasing Calluna proportion in the diet (P < 0.05). PCA of the peak areas of the metabolites identified by Pearson correlation showed distinct clustering of the three treatment groups through the combined effect of PC1 (54.0% total variance) against PC4 (1.42% total variance) (Figure 2A). Animals offered the 10% heather diet were distinguished by their PC1 score from animals on the 20 and 30% diets, whereas the latter two diet groups were distinguished by their score for PC4. The loadings plot (Figure 2B) shows this grouping was dependent on several metabolites. In particular, plasma from animals maintained on the 10% diet contained higher levels of 3-methylundecane, nicotinamide, L-isoleucine, and 4-aminobutyric acid, all of which had positive PC1 loadings. The grouping of the 20 and 30% samples corresponded to the concentrations of methoxymethylcyclopropane and N,N-diethylN′-methyl-1,2-ethanediamine, relative concentrations of which were higher in samples from the 20% heather diet. Samples from animals on the 30% heather diet also had relatively higher amounts of L-valine and malonic acid (as indicated by PCA eigenvectors). Similarly, HCA of PC1 and PC4 scores by Euclidean distance separated the samples into two clusters (Figure 2C), with one cluster comprising three of the 10% heather samples. The second cluster was subdivided further into two clusters with all of the 20 and 30% heather samples correctly grouped. Analysis of Feces Samples. Analysis of the GC-MS data from analysis of the sheep feces samples by PCA and HCA failed to distinguish between animals on each of the three diets. Pearson’s correlation was carried out to identify fecal metabolite(s) that were strongly correlated with diet. Malic acid (R ) 0.63) was found to be positively correlated (P < 0.05), whereas 2-butyl2-ethyloxazolidine (R ) -0.69) was inversely related to diet (P < 0.05). Due to the small number of variables identified, it was not possible to conduct PCA or HCA on the data. Analysis of Plant Samples. Pearson correlation was used to identify those metabolites present in samples of the mixed diets that correlated most with the amount of heather present in the diet. Results indicated that the concentrations of 5-tert-butyl2-thiophenecarboxylic acid (R ) 0.99) and 2-imino-4(5H)thiazolone (R ) 0.99) in the diet mix were most dependent on the amount of heather in the diet (P < 0.05). In contrast, nicotinic acid (R ) -0.99), malic acid (R ) -0.99), xylose (R ) -0.99), and 2-aminoadipic acid (R ) -0.99) were all negatively correlated (P < 0.05). Relationship of Metabolites in Blood Plasma, Feces, and Plant Samples to Botanical Dietary Composition. Analysis of Plasma Samples. Correlation analysis of the first 10 PC scores of the metabolite abundances in the plasma samples against the amount of each plant subgroup eaten showed a negative correlation between the amounts of Calluna (R ) -0.61, P < 0.05) and Erica (R ) -0.62, P < 0.05) consumed and PC1.

1132

J. Agric. Food Chem., Vol. 55, No. 4, 2007

Parveen et al.

Figure 1. Plots of PC1 and PC4 scores (A) and PC1 and PC4 loadings (B) (standard error bars represent standard error of means for each of the three diet groups) of plasma samples from sheep maintained on diets of 10% heather (b), 20% heather (9), and 30% heather (2).

Consistent with this trend, the total amount of heather in the diet, that is, Calluna and Erica combined (R ) -0.65, P < 0.05), was also negatively correlated with PC1. In contrast, the amount of Molinia in the diet was positively related to PC1 (R ) 0.58, P < 0.05). Similarly, the 12 metabolites identified previously by Pearson correlation analysis as being most significantly correlated to the amount of heather in the diet correlated strongly with the first principal component from the plasma PCA. Statistical analysis by ANOVA showed significant differences in the mean concentrations of L-valine (P < 0.01), 4-aminobutyric acid (P < 0.05), galactaric acid (P < 0.05), succinic acid (P < 0.05), 2-propenoic acid (P < 0.05), and methoxymethylcyclopropane (P < 0.05) in plasma samples from sheep on each of the dietary treatments. Samples from animals on the 20 and 30% heather diets contained significantly higher concentrations of these compounds than samples from animals on the 10% diet (Table 2). Similarly, there were significant

between-treatment differences in the concentrations of L-alanine (P < 0.01) and malonic acid (P < 0.001) as the proportion of heather in the diet increased. Differences in the mean concentrations of several other compounds were observed that fell below 95% level of confidence. For example, concentrations of L-isoleucine and N,N-diethyl-N′-methyl-1,2-ethanediamine were higher in samples from animals on the 20 and 30% diets compared to the 10% diet, whereas the concentration of nicotinamide decreased with increased proportion of heather in the diet. Analysis of Feces Samples. Examination of the correlation between PCs 1-10 for the GC-MS fecal data against diet composition showed that the amount of Calluna (R ) -0.54, P < 0.1) and the total content of heather (R ) -0.55, P < 0.1) in the diet were negatively correlated with PC4. In contrast, the amount of Molinia (R ) 0.68, P < 0.05) consumed was positively related to PC4. Furthermore, Festuca (P < 0.001) and Carex (P < 0.05) showed strong positive correlations,

Metabolite Profiling of Sheep and Their Diets

J. Agric. Food Chem., Vol. 55, No. 4, 2007

1133

Figure 2. Plots of PC1 and PC4 scores (A) and PC1 and PC4 loadings (B) (standard error bars represent standard error of means for each of the three diet groups) of plasma samples (normalized peak areas that correlated strongly with diet) from sheep maintained on diets of 10% heather (b), 20% heather (9), and 30% heather (2). (C) HCA of PC1 and PC4 scores for sheep plasma maintained on one of the three diets.

1134

J. Agric. Food Chem., Vol. 55, No. 4, 2007

Parveen et al.

Table 2. Mean Concentrations of Metabolites (Relative Abundance) in Plasma and Feces Samples from Sheep Maintained on One of Three Diets treatment (heather in diet) plasma metabolites L-alanine L-isoleucine L-valine 4-aminobutyric acid galactaric acid malonic acid 2-propenoic acid succinic acid methoxymethylcyclopropane 3-methylundecane N,N-diethyl-N′-methyl-1,2-ethanediamine nicotinamide fecal metabolites 2-butyl-2-ethyloxazolidine malic acid a

10%

20%

30%

SED

significancea

3.46 0.209 5.6 3.75 0.0042 9.0 0.124 0.121 3.5 0.398 5.4 0.0456

13.87 0.385 23.6 6.71 0.0131 24.7 0.295 0.259 9.6 0.759 12.1 0.0053

11.19 0.445 23.0 7.30 0.0361 32.6 0.327 0.329 10.6 0.835 12.5 0.0058

2.24 0.092 5.20 1.19 0.0097 5.8 0.072 0.064 2.6 0.145 2.9 0.0170